Part 2: Playing in the Right Key

Translating human responses into structured and usable data }

Once we got one thing working — a clean, natural conversation to collect CBT-I sleep data — we didn’t stop there.

We started throwing harder problems at the model. Ask this patient if they’re experiencing side effects from a new medication.

Some side effects are expected. Others are subtle.

Ask this one to report their A1C levels — before and after meals, across a few days.

Track symptoms, behavior, pain, mood — and do it all without sounding like a form.

And surprisingly, it worked.

The model engaged. The responses came in. People stayed with it.

But we ran into a wall.

The conversations were good. Sometimes great. But the data?

It was messy. Unique. Impossible to compare.

One person might say, “I guess I’ve been feeling kind of jittery?”

Another might write, “No big side effects, just weird dreams and some tightness in my chest.”

Interesting? Yes

Usable? No

We realized that if we wanted Genversation to matter — not just in theory, but in practice — we had to close the loop.

Not by limiting what people could say, but by shaping how it was captured.

We weren’t trying to reinvent the way clinical data is structured or reported.

We were inventing a new way to capture and extract it — through dialogue instead of forms. Through natural language feedback, instead of checkboxes.

So we turned to standards: Likert scales. Validated tools. Instruments already trusted in healthcare.

Not because they were elegant — but because they were accepted.

It’s like walking into a studio with a brilliant melody scribbled backwards in crayon. The idea might be amazing and beautiful. But if no one can read it or play it, no one will care.

So we trained the model not just to speak human, but to document like a clinician.

To take something as unstructured as a late-night text —and turn it into something that fits into a care plan.

Genversation now have three clear job functions:

1. Feel natural to a person

2. Extract useful feedback

3. Generate usable data and analytics

That’s how it will work and scale in the real world.

READ Part 3: Building Mastery